A system with multiple cooperating unmanned aerial vehicles (multi-UAVs) can use its advantages to accomplish complicated tasks. Recent developments in deep reinforcement learning (DRL) offer good prospects for decision-making for multi-UAV systems. However, the safety and training efficiencies of DRL still need to be improved before practical use. This study presents a transfer-safe soft actor-critic (TSSAC) for multi-UAV decision-making. Decision-making by each UAV is modeled with a constrained Markov decision process (CMDP), in which safety is constrained to maximize the return. The soft actor-critic-Lagrangian (SAC-Lagrangian) algorithm is combined with a modified Lagrangian multiplier in the CMDP model. Moreover, parameter-based transfer learning is used to enable cooperative and efficient training of the tasks to the multi-UAVs. Simulation experiments indicate that the proposed method can improve the safety and training efficiencies and allow the UAVs to adapt to a dynamic scenario.

No abstract

This paper proposes an algorithm for missile manoeuvring based on a hierarchical proximal policy optimization (PPO) reinforcement learning algorithm, which enables a missile to guide to a target and evade an interceptor at the same time. Based on the idea of task hierarchy, the agent has a two-layer structure, in which low-level agents control basic actions and are controlled by a high-level agent. The low level has two agents called a guidance agent and an evasion agent, which are trained in simple scenarios and embedded in the high-level agent. The high level has a policy selector agent, which chooses one of the low-level agents to activate at each decision moment. The reward functions for each agent are different, considering the guidance accuracy, flight time, and energy consumption metrics, as well as a field-of-view constraint. Simulation shows that the PPO algorithm without a hierarchical structure cannot complete the task, while the hierarchical PPO algorithm has a 100% success rate on a test dataset. The agent shows good adaptability and strong robustness to the second-order lag of autopilot and measurement noises. Compared with a traditional guidance law, the reinforcement learning guidance law has satisfactory guidance accuracy and significant advantages in average time and average energy consumption.